Three mechanisms have been proposed to underlie reciprocation of social behaviors in gregarious animals: ‘calculated reciprocity’, ‘emotional bookkeeping’ and ‘symmetry-based reciprocity’. Among these explanations, emotional book-keeping has received the broadest support from experimental and observational studies. On the other hand, three individual-based models have shown that reciprocation may emerge via ‘symmetry-based reciprocity’, ‘emotional bookkeeping’, or a combination of both mechanisms. Here we use these three models to assess their relative fit with empirical data on reciprocation and social network structure across different groups and species of macaques. We collected grooming data from 14 groups and 8 macaque species and simulated each group in each model. We analyzed and quantitatively compared social network metrics of the empirical and the models’ grooming networks. The three models captured fairly well the features of observed networks, and fitted data from wild groups better than captive ones. The emotional bookkeeping model seemed on average to fit slightly better the social networks metrics observed in empirical data, but failed to reproduce some grooming patterns. The symmetry-based models, on the other hand, fitted better other network parameters (e.g. modularity). No model generally fitted the data better than the others, and the fit with some metrics (e.g. modularity, centralization index) was low even after optimization. Thus, our analyses indicate that in the models social interactions may be simpler than in reality and models may miss social processes (e.g. third-party awareness).